[all-commits] [llvm/llvm-project] e858b1: Revert "[SLP]Reduce number of alternate instructio...

[Hans Wennborg via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: e858b10917046b83234bf1931485df414fcded3c
      https://github.com/llvm/llvm-project/commit/e858b10917046b83234bf1931485df414fcded3c
  Author: Hans Wennborg <hans at chromium.org>
  Date:   2025-03-11 (Tue, 11 Mar 2025)

  Changed paths:
    M llvm/include/llvm/Analysis/TargetTransformInfo.h
    M llvm/include/llvm/Analysis/TargetTransformInfoImpl.h
    M llvm/lib/Analysis/TargetTransformInfo.cpp
    M llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
    M llvm/lib/Target/X86/X86TargetTransformInfo.h
    M llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
    M llvm/test/CodeGen/AMDGPU/materialize-frame-index-sgpr.gfx10.ll
    M llvm/test/CodeGen/AMDGPU/materialize-frame-index-sgpr.ll
    M llvm/test/CodeGen/AMDGPU/schedule-amdgpu-tracker-physreg-crash.ll
    M llvm/test/Transforms/SLPVectorizer/AArch64/tsc-s116.ll
    M llvm/test/Transforms/SLPVectorizer/RISCV/complex-loads.ll
    M llvm/test/Transforms/SLPVectorizer/RISCV/reductions.ll
    M llvm/test/Transforms/SLPVectorizer/X86/alternate-cast-inseltpoison.ll
    M llvm/test/Transforms/SLPVectorizer/X86/alternate-cast.ll
    M llvm/test/Transforms/SLPVectorizer/X86/alternate-fp-inseltpoison.ll
    M llvm/test/Transforms/SLPVectorizer/X86/alternate-fp.ll
    M llvm/test/Transforms/SLPVectorizer/X86/alternate-int-inseltpoison.ll
    M llvm/test/Transforms/SLPVectorizer/X86/alternate-int.ll
    M llvm/test/Transforms/SLPVectorizer/X86/buildvector-schedule-for-subvector.ll
    M llvm/test/Transforms/SLPVectorizer/X86/gathered-shuffle-resized.ll
    M llvm/test/Transforms/SLPVectorizer/X86/long-full-reg-stores.ll
    M llvm/test/Transforms/SLPVectorizer/X86/lookahead.ll
    M llvm/test/Transforms/SLPVectorizer/X86/matched-shuffled-entries.ll
    M llvm/test/Transforms/SLPVectorizer/X86/non-load-reduced-as-part-of-bv.ll
    M llvm/test/Transforms/SLPVectorizer/X86/phi.ll
    M llvm/test/Transforms/SLPVectorizer/X86/reorder-phi-operand.ll
    M llvm/test/Transforms/SLPVectorizer/X86/reorder_diamond_match.ll
    M llvm/test/Transforms/SLPVectorizer/X86/same-values-sub-node-with-poisons.ll
    M llvm/test/Transforms/SLPVectorizer/X86/scatter-vectorize-reused-pointer.ll
    M llvm/test/Transforms/SLPVectorizer/X86/splat-score-adjustment.ll
    M llvm/test/Transforms/SLPVectorizer/X86/vec_list_bias-inseltpoison.ll
    M llvm/test/Transforms/SLPVectorizer/X86/vec_list_bias.ll
    M llvm/test/Transforms/SLPVectorizer/X86/vec_list_bias_external_insert_shuffled.ll
    M llvm/test/Transforms/SLPVectorizer/addsub.ll

  Log Message:
  -----------
  Revert "[SLP]Reduce number of alternate instruction, where possible"

This caused failures such as:

  Instruction does not dominate all uses!
  %29 = insertelement <8 x i64> %28, i64 %xor6.i.5, i64 6
  %17 = shufflevector <8 x i64> %29, <8 x i64> poison, <6 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6>

see comment on https://github.com/llvm/llvm-project/pull/123360

> Previous version was reviewed here https://github.com/llvm/llvm-project/pull/123360
> It is mostly the same, adjusted after graph-to-tree transformation
>
> Patch tries to remove wide alternate operations.
> Currently SLP vectorizer emits something like this:
> ```
> %0 = add i32
> %1 = sub i32
> %2 = add i32
> %3 = sub i32
> %4 = add i32
> %5 = sub i32
> %6 = add i32
> %7 = sub i32
>
> transformes to
>
> %v1 = add <8 x i32>
> %v2 = sub <8 x i32>
> %res = shuffle %v1, %v2, <0, 9, 2, 11, 4, 13, 6, 15>
> ```
> i.e. half of the results are just unused. This leads to increased
> register pressure and potentially doubles number of operations.
>
> Patch introduces SplitVectorize mode, where it splits the operations by
> opcodes and produces instead something like this:
> ```
> %v1 = add <4 x i32>
> %v2 = sub <4 x i32>
> %res = shuffle %v1, %v2, <0, 4, 1, 5, 2, 6, 3, 7>
> ```
> It allows to improve the performance by reducing number of ops. Also, it
> turns on some other improvements, like improved graph reordering.
>
> [...]

This reverts commit 9d37e61fc77d3d6de891c30630f1c0227522031d as well as
the follow-up commit 72bb0a9a9c6fdde43e1e191f2dc0d5d2d46aff4e.



To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications